Modeling of physical properties of apple slices (Golab variety) using artificial neural networks

Abstract

 Apple is one of the most popular fruits and of high economic value.  Sorting and grading of apple is needed for the fruit to be presented to local and foreign markets.  A study of apple physical properties therefore is imperative.  In this work, some physical properties of apples (Golab variety) such as main diameter, mass, volume and fruit density were determined and relation between mass and other parameters were modeled by using artificial neural networks.  In this study, we used Feed-Forward Back Propagation (FFBP) network with training algorithms, Levenberg-Marquard and Momentum.  The results show that Levenberg-Marquard algorithm give better result than Momentum algorithm do, and Feed-Forward Back Propagation (FFBP) network with topology 3-6-4-1, 3-6-1, 3-4-2-2-1 and 3-6-6-1; and Levenberg-Marquard algorithm could predict relation between mass and other parameters with error percentages 0.999999, 0.999999, 0.999999 and 0.999999; and mean square error 0.000078, 0.000118, 0.000158 and 0.000194.

Keywords: apple (Golab variety), artificial neural network, Feed-Forward Back Propagation, Levenberg-Marquard algorithm, Momentum algorithm, physical properties